US 6901514 B1 Abstract A method for embedding a watermark into content is disclosed. The content contains content samples. The method including the steps of: receiving the content, creating a continuous watermark sequence from the watermark, and for each content sample in a first predetermined order: calculating a sample mean, calculating a sample variance, and normalizing the content. Further steps include generating a set of content coefficients from the content, generating a set of watermark coefficients from the watermark sequence, embedding the watermark in the content by adjusting the amplitude of the watermark coefficients so that the distortion between the content coefficients and the associated watermark coefficients are minimized using a secret mapping function, and outputting the content. The mapping functions may be controlled by a key-dependent random sequence to protect the watermarks.
Claims(23) 1. A method for embedding a watermark into content, said content containing content samples, comprising the steps of:
(a) receiving said content;
(b) creating a continuous watermark sequence from said watermark;
(c) for each content sample in a first predetermined order:
(i) calculating a sample mean;
(ii) calculating a sample variance; and
(iii) normalizing said content;
(d) generating a set of content coefficients from said content;
(e) generating a set of watermark coefficients from said watermark sequence;
(f) embedding said watermark in said content by adjusting the amplitude of said watermark coefficients so that the distortion between the content coefficients and the associated watermark coefficients are minimized using a secret mapping function; and
(g) outputting said content.
2. The method according to
3. The method according to
4. The method according to
5. The method according to
6. The method according to
7. The method according to
8. The method according to
9. The method according to
10. The method according to
11. The method according to
12. The method according to
13. The method according to
14. The method according to
15. The method according to
16. The method according to
17. The method according to
18. The method according to
19. The method according to
20. A method for extracting a watermark sequence from watermarked content comprising the steps of:
(a) receiving watermarked content comprising received coefficients;
(b) generating an estimated watermark determined by received coefficients, and a mapping function;
(c) generating a watermark sequence using a correlation function, said correlation function using the watermarked content, the estimated watermark, a scaling factor, and a weight ing factor per a predetermined equation; and
(d) outputting the watermark sequence.
21. An apparatus for extracting a watermark sequence from watermarked content comprising:
(a) a noise source;
(b) a key dependent sequencer;
(c) a mapping function having parameters, at least one of said parameters receiving input from said key dependent sequencer;
(d) a watermark estimator, said watermark estimator generating a watermark estimate from the watermarked content, and the mapping function
(e) a scale factor;
(f) a weight factor; and
(g) a correlator, said correlator generating the watermark sequence from the watermarked content, the scale factor, the weight factor and the water mark estimate.
22. A n apparatus for embedding a watermark data into content including:
(a) a content preprocessor, said content preprocessor further including:
(i) a mean calculator; and
(ii) a variance calculator;
(b) a content coefficient generator for generating content coefficients from the preprocessed content;
(c) a watermark sequence generator for generating a watermark sequence from the watermark data;
(d) a watermark coefficient generator for generating watermark coefficients from the watermark sequence; and
(e) a watermark inserter for generating watermarked content.
23. An apparatus according to
(a) a key dependent sequencer,
(b) a secret mapping function device, said secret mapping function device receiving input from said key dependent sequencer; and
(c) a coefficient modifier for generating watermarked content by adjusting the amplitude of the watermark coefficients so that the distortion between the content coefficients and the associated watermark coefficients are minimized using the secret mapping function device.
Description The present application claims the benefit of provisional patent application Ser. No. 60/136,961 to Iu et al., filed on Jun. 1, 1999, entitled “Secure Oblivious Watermarking using Key-Dependent Mapping Functions”, which is hereby incorporated by reference. This invention relates generally to data protection, and more particularly to aspects of a novel digital watermark system and methodology for multimedia content, such as audio, video, text, still images, computer graphics, and software. A watermark is an imperceptible or at least difficult to perceive signal embedded into multimedia content such as audio, video, text, still images, computer graphics, or software. The watermark conveys some useful information without disturbing or degrading the presentation of the content in a way that is noticeable or objectionable. Watermarking techniques play an important role in protecting copyright ownership of digital contents including images, audio, and video. Watermarks may be used to identify the original owner of the content, to trace where pirate copies of the content come from (fingerprinting), and to determine royalty payments by monitoring the number of times content has been used. Watermarks may also be used to authenticate original content and to locate change in a corrupted or altered copy of the content. In order to encourage copyright owners to use watermarking schemes, four basic and conflicting requirements should be met. Firstly, the distortion introduced by embedding the watermarks into content should be unperceivable by regular users. Secondly, the watermarks should be secure so that they are hard to be modified or removed by the pirates. Thirdly, the watermarks should be robust against intentional attacks, ranging from simple content manipulation such as cropping, to common image processing techniques, such as filtering and compression. Lastly, the overall cost of using watermarking should not be expensive. Watermarking schemes may be categorized as non-oblivious or oblivious, depending on whether the original content is available or not. Oblivious watermarking may be defined as a watermarking scheme in which the original image is not available during watermarking decoding. Non-oblivious image watermarking schemes in general may be more robust due to the accessibility of the original image because image distortions caused by image processing, transmission, or intentional attacks may be compensated for using the original image. Also, the interference between the original image and the watermarks during watermark decoding may be removed by using the difference of the watermarked and original images. However, for many applications, such as copy and playback control, and copyright protection, the requirement of accessing the original image is simply not practical. This may make oblivious watermarking the only choice. Watermarks may be embedded in the pixel or the transform domains. Two papers which discuss and compare different methodologies and watermarking schemes include “A fair benchmark for image watermarking systems”, by M. Kutter and F. A. P. Petitcolas, (SPIE Electronic Imaging'99: Security and Watermarking of Multimedia Contents, vol. 3657, January 1999), and “Comparing robustness of watermarking techniques” by J. Fridrich and M. Goljan (SPIE Electronic Imaging'99: Security and Watermarking of Multimedia Contents, vol. 3657, January 1999). Proposed transforms include DCT, DFT, LOT, wavelets, Hadamard transform and key-dependent transforms. The watermark signal in a transform domain may usually be related to that in the pixel domain by a linear transformation, if the transform itself is linear. However, the analysis may be applied to pixel-based approaches as well. Human visual models have been used to adjust watermark strength so that embedded watermarks may be invisible. Spread-spectrum techniques are widely used by most oblivious watermarking approaches. When extracting the watermark message, these methods may rely on the watermark information embedded in the middle frequencies, although the noise-like watermark signal may also be embedded in the low and the high frequencies. The watermark information in the high frequencies may be easily removed using low-pass filtering and JPEG compression, and humans may be able to tolerate high distortion there. For low frequencies, watermark signals may have a high interference with the image itself. Note that the energy of a typical image may be concentrated in the lower frequencies. For non-oblivious watermarking, adding watermarks in the low frequencies has been shown to have some advantages in a paper entitled “A review of watermarking and the importance of perceptual modeling”, by I. Cox and M. Miller, Proc. of the SPIE Human Vision and Electronic Imaging, vol. 3016, pp. 92-99, February 1997. More watermark messages may be sent while the noise level of the image does not increase. Watermarks in the low frequencies in general may be more robust than that in the middle frequencies, with respect to image distortions that have low-pass characteristics, such as filtering. Examples of nonlinear filtering, may include median filting, lossy compression filtering, and adaptive Wiener filtering. Watermarks in the low frequencies may also be less sensitive to small geometric distortions (e.g., rotation, shifting, and scaling). Therefore, seeking oblivious watermark schemes unitizing the low frequencies and distortion compensation techniques without the original image have become two active research topics. Several watermark attack and counterattack methods have been proposed. To overcome a geometric attack, small blocks of a corrupted image may be registered with an original pseudo noise signal using correlation matching. Watermarks may also be removed by capturing watermark information pixel by pixel with a sensitivity attack if a pirate has access to a device that can detect whether the content contains a watermark or not. To handle distortions without the original image, a calibration pattern may be embedded into the Fourier transform in the log-polar coordinates, so that the shift, scaling, s and rotation of the image may be compensated. Some oblivious watermarking approaches using the low frequency bands have been proposed including embedding watermark information by swapping selected transform coefficients of 8×8 DCT blocks. The robustness of this type of approach may not be high and visible distortions may be introduced. Another approach includes embedding watermark message bits into disjoint triplets of wavelet coefficients, which may be chosen according to a key-dependent random sequence. The middle coefficient may be quantized by a quantization step, what is equal to the difference of the largest and the smallest values of the triplet, divided by a fixed scale factor. This approach may not be applied to DCT coefficients since the standard deviation of the DCT coefficients in low frequencies may typically be very high. This requires a large fixed scale factor, or equivalently a small quantization step, in order for the watermark to remain invisible. Therefore, the robustness has to be compromised. Similar quantization techniques have been proposed to embed a cartoon or map image into a host image. Quantization with frequency and spatial masking to embed watermarks into DCT Yet another proposed method includes using the quantization index modulation to embed a watermark message into a host image. Message bits are used to select the pre-defined quantizers. Theoretical results for some channel models have been discussed. However, no experiments on real distortions have been reported. Watermarks inserted in the middle and high frequencies may typically be very robust with respect to noise adding, nonlinear deformations of the gray scale, (e.g., contrast/brightness adjustment, gamma correction, histogram equalization), and cropping. Since these advantages are complementary to that of low-frequency techniques, and watermarks of low and middle frequencies are embedded into disjoint portions of the spectrum, Fridrich proposed to embed both low and high frequency watermarks into the image. To decode the hidden message in the low frequencies without the original image, binary mapping may be used. A (watermark) mapping function (also called an index function) may relate the watermarked transform coefficient to the watermark itself. Although it has been shown that the watermarks may be very robust to different types of distortion, there may be a serious security problem. The watermarks may be easily removed by clustering the DCT coefficients using a histogram attack, which may search for the parameters of the mapping function. If the intensity of some of the original pixels are guessed and if the basis function is known, then the watermarks may be estimated and the general watermark system fails. To overcome the histogram and the watermark-estimation attacks, it has been proposed that some secret key-dependent basis functions could be used. Although this scheme seems to be able to achieve better robustness to different distortions and high security to attacks, it may require very high computation and a relatively large amount of storage to generate the basis functions and to find the corresponding transform coefficients. We will now discuss briefly an oblivious watermark approach, described in a paper by J. Fridrich, entitled “Combining low-frequency and spread spectrum watermarking”, Proc. SPIE Int. Symp. on Optical Science, Engineering Instrumentation, San Diego, July 1998, which uses a binary mapping function. A security problem will be disclosed using a histogram attack. The oblivious low frequency watermarking of Fridrich is described as follows. Let f During watermark decoding, the watermark may be estimated from the received DCT coefficient u Combining with mid-frequency watermarking using the spread spectrum technique, the above binary watermarking has been shown by Fredrich to be robust for many attacks. However there arises a serious security problem. Since the watermarked coefficients u Fridrich has observed this problem. He also discussed the security problem faced under the watermark-estimation attack. If the original intensity of some pixels of a watermarked image can be guessed, then the watermarks may be estimated and removed by solving a system of linear equations. To address both security problems, Fredrich proposed the use of key-dependent basis functions. He also demonstrated that his approach was quite robust to common distortions. However, his approach requires a high computation to generate the transform functions and to perform the forward or inverse transforms. To provide an alternative, the present invention will disclose a new class of mapping functions, which may require only simple operations. These mapping functions may be controlled by a secret key. To combat the watermark-estimation attacks, some counter-attacks will also be disclosed. What is needed is a simple and effective scheme to enhance the security and robustness of a low-frequency watermarking scheme that protects the watermarks by using a secret (watermark) mapping function instead of a secret transform basis function. The scheme should also reduce the interference between the watermarks and the image itself by using a key-dependent quantization function. The scheme should also be generalized so that it may be applied to pixel-domain watermarking schemes. To combat the watermark-estimation attack, a simple counterattack is also needed that that the use of key-dependent basis functions isn't needed. One advantage of the invention is that it that protects watermarks by using a secret (watermark) mapping function instead of a secret transform basis function. Another advantage of this invention is that it reduces the interference between the watermarks and the image itself by using a key-dependent quantization function. Yet a further advantage of this invention is that it is generalized so that it may be applied to pixel-domain watermarking schemes. To achieve the foregoing and other advantages, in accordance with all of the invention as embodied and broadly described herein, a method for embedding a watermark into content. The content contains content samples. The method including the steps of: receiving the content, creating a continuous watermark sequence from the watermark, for each content sample in a first predetermined order: calculating a sample mean, calculating a sample variance, and normalizing the content. Further steps include generating a set of content coefficients from the content, generating a set of watermark coefficients from the watermark sequence, embedding the watermark in the content by adjusting the amplitude of the watermark coefficients so that the distortion between the content coefficients and the associated watermark coefficients are minimized using a secret mapping function, and outputting the content. In yet a further aspect of the invention, a method for a method for embedding a watermark into content, wherein the step of embedding the watermark in the content is performed by adjusting the watermark coefficients sequentially in a second predetermined order. In yet a further aspect of the invention, an apparatus for embedding a watermark data into content including: a content preprocessor, the content preprocessor further including: a mean calculator; and a variance calculator, a content coefficient generator for generating content coefficients from the preprocessed content; a watermark sequence generator for generating a watermark sequence from the watermark data; a watermark coefficient generator for generating watermark coefficients from the watermark sequence; and a watermark inserter for generating watermarked content. The watermark inserter may further include a key dependent sequencer; a secret mapping function device, the secret mapping function device receiving input from the key dependent sequencer; and a coefficient modifier for generating watermarked content by adjusting the amplitude of the watermark coefficients so that the distortion between the content coefficients and the associated watermark coefficients are minimized using the secret mapping function device. Additional objects, advantages and novel features of the invention will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following or may be learned by practice of the invention. The objects and advantages of the invention may be realized and attained by means of the instrumentalities and combinations particularly pointed out in the appended claims. The accompanying drawings, which are incorporated in and form a part of the specification, illustrate an embodiment of the present invention and, together with the description, serve to explain the principles of the invention. The present invention is a new method for watermarking content using a novel class of secure mapping watermarking functions using key-dependent mapping functions. The binary watermarking scheme disclosed by Fredrich and discussed in the background section may be defeated by a histogram attack because the DCT coefficients may be clustered in the middle of the interval for the correct quantization step size β. The present invention overcomes this security problem by first replacing the binary watermark sequence by a continuous sequence and by using secret mapping functions for different DCT coefficients. Replacing the binary watermark sequence by a continues sequence may cause the DCT coefficients after quantization to spread out in the quantization interval, making the search for the original quantization step size β difficult. Using secret mapping functions for different DCT coefficients may make the histogram attack impossible because the mapping functions may not be known and may be changed for different DCT coefficients. In general, many functions may be used as the mapping functions. The functions may be generated by a program or retrieved from a look-up table. For the robustness concern to different distortions or attacks, it may be required that these functions are continuous or at least piecewise continuous. Otherwise, a small change of the DCT coefficients may introduce a big error in the estimated watermark. To make these mapping functions practically useful, it may be important that these functions may be rapidly computed or generated in real-time and/or they do not require a large space to store their values. The present invention may use mapping functions which preferably take a simple function form. Their parameters may be controlled by some key-dependent random sequences to offer security. Assume that the watermark sequence w From (5) and (7), it may be shown that w Since the mapping function M One more layer of security may be offered via the design of the mapping function by allowing a varying size of the quantization intervals for quantizing the DCT coefficients, resulting in the interval length preferably varying for different i, i.e. I For the saw-toothed mapping function, there may be a problem for the robustness of the watermark recovery. If the noise n This may be achieved by simply assigning q(u For the binary mapping function, its security problem may be overcome by using randomized quantization steps, i.e.
This scheme protects the watermarking system under histogram attack. Since each DCT coefficient has its own quantization step size, the attack which searches for a common step size would fail and the histogram will appear to be random. Histograms for the binary and saw-toothed mapping functions with a random sequence of β For generalized mapping functions, periodic functions with key-dependent parameters may be used as follows
Rewatermarking may be accomplished with and without quantization, where the received watermark coefficients may be described by
For the binary, saw-toothed, and triangle mapping functions in equations 4, 10, and 13, respectively, their distortions during watermark encoding and the watermark estimate error during watermark decoding are analyzed. Note that since these mapping functions are periodic, analyzing one period of each function is sufficient without loss of generality. Let Δu For the binary mapping function, if the log-magnitude of the i-th DCT coefficient after watermarking, u For the binary mapping function, if the received DCT coefficient u For the sawtoothed mapping function, from equation 10, we have
Finally, if u Comparing the saw-toothed and the triangle mapping functions, when u
One type of attack to a watermark system is to estimate the unknown watermarks from a given watermarked image f Let v Another counterattack to offend the watermark-estimation attack may be to add some noise to the image to corrupt quality of the estimation of v Fortunately, there is a simple solution for solving this problem. For the pixels whose intensity may be estimated easily, after watermarking, the pixels in the watermarked image may be replaced by the pixels in the original image. In this case, V Furthermore, some experiments have been performed for the watermark-estimation attack. It was found that the estimate of v We will now start to describe some embodiments of the present invention by referring to A simple but effective way to protect the watermarks for oblivious watermarking by using a new class of mapping functions has been disclosed. These functions may be controlled by key-dependent random sequences. The watermark encoding and decoding may only require a simple computation. A security problem of a binary mapping function may be overcome by using random quantization steps. The disclosed mapping functions may be applied to pixel-based approaches and other transform-based approaches including the key-dependent basis functions. The discussion on watermark-estimation attack indicates that this attack may be defeated easily. Thus, the use of key-dependent mapping functions may provide an alternative to build a secure and robust oblivious system instead of using the time-consuming key-dependent basis functions. The foregoing descriptions of the preferred embodiments of the present invention have been presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the invention to the precise forms disclosed, and obviously many modifications and variations are possible in light of the above teaching. The illustrated embodiments were chosen and described in order to best explain the principles of the invention and its practical application to thereby enable others skilled in the art to best utilize the invention in various embodiments and with various modifications as are suited to the particular use contemplated. For example, one skilled in the art will recognize that the present invention may be used with other types of content besides just images such as music, video, and data. It is intended that the scope of the invention be defined by the claims appended hereto. Patent Citations
Referenced by
Classifications
Legal Events
Rotate |